White House admits keyword searches canceled grants in lawsuit

The most important fact in Thakur v. Trump is not that the administration canceled a very large amount of research funding. It is that, in signed July 15, 2026 court stipulations, federal agencies admitted using keyword searches rather than individualized findings that University of California grant recipients had violated grant terms. The words doing the work included “diversity,” “gender,” “vaccine hesitancy,” “COVID-19,” “health equity,” “structural racism,” and “sexual orientation,” according to reporting that cites the stipulations directly.[1][2]

That distinction matters because a stipulation is not campaign rhetoric, not a press characterization, and not a plaintiff’s inference from circumstantial evidence. It is an agreed litigation fact. For a case now moving toward an Oct. 20, 2026 summary judgment hearing before Judge Rita Lin in the Northern District of California, the admissions narrow the factual dispute that would otherwise occupy the center of the briefing: whether the government was actually enforcing grant conditions or instead screening federally funded research for disfavored subject matter and political associations.[2][3]

What the admissions actually cover

The stipulations, as reported, do not merely say that agencies considered policy priorities when reviewing grants. They admit a screening mechanism. NIH targeted grants containing specified terms. DOE, the Department of Transportation, and EPA used similar keyword screens. NIH alone suspended or canceled more than 1,000 UC grants, including nearly 700 at UCLA.[2]

The Department of Energy stipulation is even more pointed for the equal protection claim. DOE admitted that it terminated 283 grants in states won by Kamala Harris while leaving 340 grants in states won by Donald Trump untouched, based “solely on political identity of the grant recipient’s state,” according to CalMatters’ account of the stipulation.[2]

Those are not stray administrative details. They are the sort of facts that change the shape of summary judgment. A plaintiff challenging grant terminations often has to prove motive from timing, public statements, shifting explanations, or comparator evidence. Here, at least for the admitted categories, the plaintiffs can point to the government’s own record of how grants were selected for suspension or cancellation.

Why keyword screening strengthens the First Amendment claim

A federal research grant is not a roving license to compel the government to fund every project forever. Agencies can enforce grant terms. They can terminate awards for nonperformance, fraud, failure to meet conditions, or other lawful reasons. But the constitutional lawsuit over the White House keyword search to cancel grants turns on a different proposition: whether the government may use disfavored vocabulary and research subjects as triggers for post-award cancellation.

That is why “DEI grants” is an imprecise shorthand for the legal problem. It may describe the political controversy around some grants, but it does not define the constitutional category. A study containing the word “gender” could be epidemiological, sociological, medical, historical, or methodological. A grant mentioning “vaccine hesitancy” could test public-health communication, measure access barriers, or examine misinformation. A keyword search does not by itself determine whether the recipient failed to comply with an award condition.

The plaintiffs’ First Amendment theory becomes stronger because the admitted process supplies a cleaner causal chain. If grants were identified for cancellation because they contained particular terms associated with administration-disfavored ideas, the record begins to look less like neutral grants management and more like viewpoint or subject-matter discrimination. The government may still argue that the keywords were proxies for lawful programmatic criteria, or that some awards were independently terminable under grant documents. But the stipulations make that argument harder to win on a generalized record.

The Ninth Circuit context matters here. In this litigation, the court has twice found the administration likely violated the First Amendment, in August 2025 and December 2025 rulings discussed by Eugene Volokh in May 2026.[4] Those rulings do not decide final summary judgment. They do, however, tell Judge Lin that the appellate court has already viewed the First Amendment theory as serious enough to support preliminary relief. The July stipulations then add a more developed factual record to a theory the Ninth Circuit has already found plausible.

The litigation risk for the government is not simply that the keywords are politically charged. It is that the agencies allegedly used them at the selection stage before making individualized noncompliance determinations. In summary judgment terms, that helps plaintiffs argue that unconstitutional motive is not merely one possible reading of the record. It is embedded in the screening method the agencies admitted using.

The DOE stipulation gives the equal protection theory a sharper edge

The equal protection claim does not rest on the same facts as the First Amendment claim. Keyword screening points toward discrimination by topic or viewpoint. DOE’s admitted state-based treatment points toward discrimination by political identity. That difference matters because equal protection analysis asks whether similarly situated parties were treated differently for an impermissible reason.

Comparator evidence is often messy. The government can argue that two grants differ by agency, program, principal investigator, project milestone, award date, statutory authority, or compliance history. DOE’s stipulation, as reported, removes much of that usual fog: 283 grants from Harris-won states were terminated while 340 grants in Trump-won states were left untouched, with the distinction admitted as based solely on the political identity of the grant recipient’s state.[2]

That admission is unusually direct. It does not require the court to infer political targeting from an official’s public statement or an uneven pattern of enforcement. It states the sorting rule. If the parties have stipulated that state political identity drove differential treatment, plaintiffs can argue that the constitutional violation is visible on the face of the administrative process.

The government may still contest the legal consequence. It may argue that grant recipients do not have the comparator relationship plaintiffs claim, that political geography is not being used in the constitutionally relevant sense, or that the relevant remedial inquiry differs by agency and award. But those arguments operate downstream from a damaging admitted fact. At summary judgment, the question is not whether the DOE stipulation sounds bad in public. It is whether the admitted sorting criterion leaves any material factual dispute about impermissible discrimination.

Appropriations is the harder, but still live, structural claim

The Appropriations Clause theory asks a different question: when Congress has appropriated money and agencies have awarded grants, how much room does the executive branch have to cancel those grants because later leadership dislikes the funded work or has shifted agency priorities?

On that issue, the July 17, 2026 ruling in Massachusetts v. OMB is important legal weather. Judge Talwani ruled that an “agency priorities” clause did not authorize retroactive cancellation of grants after award for a 24-state coalition challenging federal funding cuts.[5][6] That does not bind Judge Lin in the Northern District of California, but it puts pressure on the government’s statutory fallback argument.

The keyword stipulations matter to the appropriations theory because they make the government’s explanation look less like ordinary administration of congressionally funded programs and more like post-award reallocation by ideological screen. If an agency terminates a grant after finding that a recipient missed deliverables, misused funds, or failed a condition in the notice of award, the Appropriations Clause claim is weaker. If the agency cancels awarded funding because the project contains disfavored terms, the structural objection becomes more concrete.

Still, this is the claim where caution is most warranted. The research record flags unresolved tension around separation-of-powers theories after the D.C. Circuit’s Global Health Council decision addressing an Impoundment Control Act private-right-of-action issue and related Supreme Court stay activity. Those developments do not erase the plaintiffs’ appropriations argument in Thakur, but they make it harder to treat the structural claim as settled simply because another district court rejected the “agency priorities” rationale.

For litigators watching the case, the useful distinction is this: the stipulations are most immediately powerful on motive and classification. They are also relevant to appropriations, but the separation-of-powers remedy may depend on appellate doctrine that is still in motion.

The NEH decision is the closest parallel, not a perfect match

The nearest related case is Authors Guild v. NEH, where Judge Colleen McMahon ruled on May 7, 2026 that DOGE’s use of ChatGPT to terminate more than 1,400 National Endowment for the Humanities grants was unconstitutional.[7] The case is worth comparing because it involves automated or AI-assisted grant targeting, large-scale termination, and a viewpoint-discrimination theory in the federal funding context.

The comparison should stay disciplined. Authors Guild involved NEH grants and DOGE’s use of ChatGPT. Thakur involves UC research grants and admitted keyword screening by multiple agencies, with one source reporting that the NIH stipulation referenced possible AI use by DOGE.[2] The existing record described in the research materials supports the keyword-search point more securely than any broader conclusion about AI use across agencies.

Even with that caveat, the NEH ruling helps frame the constitutional stakes. Courts are being asked to decide whether government grant portfolios can be batch-processed for disfavored ideas after awards have been made. The technology may vary: a keyword list, an automated model, a spreadsheet sorted by state politics. The common legal pressure point is the same: the government must explain why its sorting mechanism is lawful rather than viewpoint-based or politically discriminatory.

What the stipulations do not prove

The stipulations give plaintiffs a strong evidentiary platform, but they do not decide every issue in the case. They do not establish the full remedial scope. They do not determine whether every canceled grant must be reinstated. They do not eliminate agency-specific defenses. They do not prevent the government from arguing that some grants had independent defects or that some plaintiffs lack the right vehicle for particular separation-of-powers claims.

They also do not end the appeal. A district court summary judgment ruling in October would almost certainly be followed by further appellate activity if it materially constrains the administration’s grant-termination program. The Ninth Circuit’s earlier First Amendment rulings make the plaintiffs’ path stronger within that circuit, but the appropriations and impoundment questions are developing unevenly across courts.

Nor should EPA settlement talks be overread. CalMatters reports that EPA is in settlement talks with plaintiffs, which may indicate fractures in the government’s unified defense posture.[2] But settlement discussions are not admissions of liability. They are useful to watch because they may reveal whether some agencies view their own records as harder to defend than others.

The October hearing now turns on a narrower record

The July stipulations make the October summary judgment hearing less about discovering what happened and more about assigning constitutional significance to what the government has already admitted. That is a meaningful shift. Plaintiffs no longer have to build the entire case from public statements, budget politics, or suspicious timing. They can start with agency admissions about keywords, grant counts, and political-state distinctions.

For the First Amendment claim, watch whether Judge Lin treats keyword screening as direct evidence of viewpoint discrimination or as an administratively crude but legally permissible way to identify grants inconsistent with current priorities. For equal protection, watch whether the DOE Harris-state versus Trump-state distinction is treated as dispositive of impermissible political discrimination or whether the court requires a more granular comparator analysis. For appropriations, watch whether Massachusetts v. OMB supplies persuasive force against the “agency priorities” defense despite unresolved appellate tension.

The strongest prediction the record supports is modest but important: the government’s signed stipulations give the UC plaintiffs an unusually favorable summary judgment record on viewpoint discrimination and politically discriminatory treatment. They also strengthen the appropriations claim, though that theory remains exposed to broader separation-of-powers uncertainty. The case to watch in October is not whether “DEI grants” are popular or unpopular. It is whether the federal government can cancel appropriated, awarded research funding because a search string found the wrong words.

References

1. White House admits it used keywords to kill billions worth of California research grants — AP News, July 21, 2026
2. Did the White House illegally cancel CA research grants? — CalMatters, July 2026
3. Thakur v. Trump – Defending the Rule of Law — Farella Braun + Martel
4. First Amendment Likely Precludes Trump Administration's Canceling DEI-Promoting Contracts — Reason/Volokh Conspiracy, May 26, 2026
5. AG Campbell Wins Ruling Protecting Billions In Critical Federal Funding — Mass.gov
6. US Judge Says Trump Administration Cannot Revoke Grants That Conflict With Its Priorities — Reuters via U.S. News, July 17, 2026
7. Authors Guild Plaintiffs Win Case Against DOGE — Authors Guild, May 7, 2026